Image registration based on Modified Complement Weighted Sum of Minimal Distance R. Kokila #1 , P. Thangavel *2 # Department of Computer Science, Jyoti Nivas College Autonomous, Karnataka, India. 1 kok.oc25@gmail.com ∗ Professor and Head, Department of Computer Science, University of Madras, Guindy Campus, Chennai - 600025, India. 2 thangavelp@yahoo.com Abstract—We propose a novel distance measure called Modified Complement Weighted Sum of Minimal Distance (MCWSMD) for image registration. We have conducted ex- periments by adding different types of noise on the grayscale images, medical images and on the face dataset. We further evaluated performance on synthetic images and then compared the results with other existing methods and found that MCWSMD outperforms the other distance based methods. Index Terms—Image registration, Distance measure, Face recognition, Set distance I. I NTRODUCTION Image registration is the process of aligning two images that share common visual information that can be taken at different times or using different sensors or at different geometric view points. It is an essential step in the image processing application. The main objective of the image registration is to find the geometric transformation that maps the source image into the target image. In this paper we study the registration problem with reference to translation and rotation of the images, using distance measures. Hatimi et al. [1] proposed an approach to track the moving object using a fixed cammera. Daway et al. [2] proposed a new strategy to precise pupil detection. Edy et al. [3] proposed a model of face recognition utilizing the asymmetrical half-join method for joining two face images from left and right lens from a stereo vision camera. The distance measure is one of the important tools to determine the correspondence between the images in order to quantify the accuracy of the image registration. Researchers have proposed a number of distance measures for different applications and problems[4]-[19]. The Hausdorff distance and its modifications are used for image registration and related problems [4]-[5], [7]-[18]. Barrow et al. [4] have proposed parametric correspondence and chamfer matching for image matching. Hierarchical chamfer matching was developed by Borgefors [5] and it was used widely in various applications of image processing. Huttenlocher et al. [6] have presented algorithms for computing the Hausdorff distance between a binary image and a model. Dubussion and Jain [7] presented a modified Hausdorff distance (MHD) based on the average distance value for object matching. Eiter and Mannila [8] introduced sum of minimal distance and it is very much useful in various applications such as computational geometry, updating or changing theories, philosophy of science and in machine learning. Object location using Hausdorff distance was presented by Rucklidge [9]. Jesorsky et al. [10] presented a face detection system using Hausdorff distance. Kwon et al. [11] proposed hierarchical Hausdorff distance matching algorithms using pyramidal structures. Weighted Hausdorff distance for word image matching was proposed by Lu et al. [12]. A new Hausdorff distance measure was proposed by Zhao et al. [13] for gray images that have a few pixel values. Zwang et al. [14] have used generalised and mean Hausdorff distance for character recognition. Zachow et al. [15] have used Hausdorff distance measure in computer-assisted planning for improved surgical preparation. Espinace et al. [16] have used Modified Hausdorff Distance in robot movements by finding the distance between the expected line segments the robot should sense and the line segments extracted from actual measurements using a range finder. Ji et al. [17] proposed Hausdorff distance based matching for outdoor mobile robot localisation. Recently ´ Curi´ c et al. [18] proposed complement weighted sum of minimal distance in which they combined good properties of sum of minimal distance and included the information from the complement of sets. This distance measure performs well for the binary images. When complement weighted sum of minimal distance is applied to register the grayscale and medical images, the performance is low. In order to overcome the problem, we proposed a new distance measure and use binary images, grayscale images, medical images and also images on a face dataset to test its performance in the image registration. The objective of proposed distance measure to provide an efficient way to register the image even in the presence of noise. In this paper we propose a novel measure called Modified complement weighted sum of minimal distance (MCWSMD) and compared it with existing measures. The proposed distance measure is able to determine the correspondence among the images. The results are compared with ´ Curi´ c et al. [18] results. This distance measure is developed in order to improve the performance of registration of translated and rotated images. This paper is organised as follow as: the basic definition on Hausdorff distance is presented in Section II. We define Modi- International Journal of Computer Science and Information Security (IJCSIS), Vol. 18, No. 9, September 2020 https://doi.org/10.5281/zenodo.4159219 46 https://sites.google.com/site/ijcsis/ ISSN 1947-5500